Economic & Labour Market Review | Vol 1 | No 3 | March 2007

Methods e x p l a i n e d
Methods explained is a quarterly series of short articles explaining statistical issues and methodologies relevant to ONS and other data in a
simple, non-technical way. As well as defining the topic areas, the notes explain when, why and how these methodologies are used. Where
relevant, the reader is also pointed to further sources of information.

Index numbers
Peter Goodridge
Office for National Statistics

SUMMARY

Many of the statistics produced by the Office for National Statistics,

particularly economic statistics, are published in the form of
indices. However, there are a number of different forms of indices
and this article attempts to explain the subtle differences in the
methodologies used to construct them, and also factors that feed
into the choice of which type of index to use. Hypothetical examples
are provided to illustrate the effects of using different index forms.

ndex numbers are a common and convenient way of expressing

time-series variables. They are useful because the movement
from one period to the next is expressed as the percentage of the
value in the base period, the choice of the base period depending on
the type of index used. The concept of the base period is discussed
later. This article summarises the common types of indices used,
particularly those used by the Office for National Statistics (ONS),
and the differences between them. Some common variables
expressed in index form are capital, labour, output, prices, earnings
and productivity.
In Table 1 below is a simple example of an index. Say we are
interested in a variable X and its evolution over time. Then the index
is simply the growth in X multiplied by the value of the index in the
previous period. That is:
I(t) =

X(t)

(X )I
(t1)

(1)

(t1)

or, for the year 2003, the index is equal to (37/34)*113.33 = 123.33.
The informed reader will notice that the above index has been
chained. The difference between chained and non-chained indices is
discussed below.

Table 1
Simple example
Time (t)
2000
2001
2002
2003
2004
2005
2006

54

Office for National Statistics

It should be noted that, when constructing an index, the variables

in question need to be in the same unit, or converted into the same
unit. Also the index in the above example was made up of just one
input variable. If the variables used to create the index are in different
categories, then these will need to be weighted together to form the
aggregate index, based on their relative importance. Weighting is
discussed below in reference to Laspeyres indices.

Laspeyres
One of the most common forms of indices used at the ONS is the
Laspeyres. The main feature of the Laspeyres is that the weights
used are taken from the base period. For example, if we consider
a price index with a base t-1, then the prices will be aggregated for
all periods using weights from that period. However, if the index is
chain-linked, then the weight will be taken from the previous period.
Chain-linking simply means updating the weights so that, for each
period, the base used is the weight from the previous period. ONS
output (GDP or GVA) measures are now calculated as chain-linked
Laspeyres indices and are referred to as chained volume measures
(CVM). For a fuller discussion of chain-linking in relation to
National Accounts output measures, see Robjohns (2006).
Below are two simple examples of a weighted price index consisting
of two goods, X and Y. The weights used are the quantities of the
goods consumed and so represent their relative importance. The
price is multiplied by the quantity, to obtain expenditure, and the
change is expressed as a percentage of the expenditure in the base
period. Equation 2 below shows a standard Laspeyres price index;
however, the principle is true for any type of index, not just prices,
that is the weight is taken from the base period. For instance, for
a quantity index, the quantity in the numerator term would be
updated rather than the price, and the denominator would remain
the same.

L
t, 0

Index (I)

30
33
34
37
35
36
40

100
110
113
123
117
120
133

P .Q
P .Q
i

i,t

i,0

i, 0

i, 0

100

(2)

Table 2 presents a Laspeyres index using a base year of 2000, and

Table 3 shows the effect of chain-linking the index.
As can be seen above, chain-linking has the strongest impact on the
index when there is a relative shift in composition of the variable.
In this case, there has been a relative shift toward consumption of

Economic & Labour Market Review | Vol 1 | No 3 | March 2007

Methods explained

Table 2
Example of a weighted Laspeyres index

Expenditure
Expenditure: in base period:
P0*Q0
Time (t) P(X)
P(Y)
Q(X)
Q(Y)
Pt*Q0
2000
2001
2002
2003
2004
2005
2006

10
14
19
28
40
44
56

5
7
6
7
8
9
11

30
32
38
43
47
53
60

60
60
840
600
55
930
600
52
1,260
600
50
1,680
600
50
1,860
600
45
2,340
600

Index:
base
(2000)
100
140
155
210
280
310
390

Table 3
Example of a weighted chain-linked Laspeyres index

Expenditure
Expenditure: in base period:
Pt-1*Qt-1
Time (t) P(X)
P(Y)
Q(X)
Q(Y)
Pt*Qt-1
2000
2001
2002
2003
2004
2005
2006

10
14
19
28
40
44
56

5
7
6
7
8
9
11

30
32
38
43
47
53
60

60
60
840
600
55
968
868
52
1,449
1,052
50
2,136
1,568
50
2,518
2,280
45
3,518
2,782

Index:
chainlinked
100
140
156
215
293
324
409

good X, the more expensive good, causing the chain-linked index

to rise by a greater amount than the standard Laspeyres, since the
price of good X has risen more than that of good Y. This is obviously
relevant to a number of areas, for instance the changing industrial
composition of the economy. Therefore, chain-linking is a way
of improving the methodology because it takes account of more
information and so provides a more accurate measure. However, the
ability to chain-link does depend on the timeliness of the data used
for the weights.
The first main benefit of chain-linking is that new items can be
added to the basket every year. If the index is non-chained, new
items can only be added to the base year. The second benefit is that
by chaining the series, the comparison is with the previous year,
rather than the base year. This is obviously relevant when most of the
interest is in the annual change such as with prices, output and most
economic indicators.
Another benefit of chain-linking is that it removes substitution bias.
This is a problem that is encountered when there are large shifts in
both the weight and in the actual variable that is being indexed. This
was encountered in the US with data for computer hardware, but
also applies more generally to all technology goods. Prices for such
goods have reduced dramatically, causing the quantities consumed
to increase. However, when the output index is unchained, it means
that the index is being weighted on prices from the base year, when
current prices are far removed from those in the base year, that is,
the weights used are too high. Therefore, when the index is rebased,
which in the US was every five years, the rebasing has the effect of
reducing the index, causing large revisions. Chain-linking removes
this problem.
The retail prices index (RPI) is an annually chain-linked Laspeyres
index.1 Other data produced as Laspeyres indices are the qualityadjusted labour input measure (QALI), the volume index of capital

services (VICS), both produced by the ONS, and also the house price
index, produced by the Department for Communities and Local
Government (DCLG).

Paasche
The main difference between Laspeyres and Paasche indices is that,
with the Paasche, the weights are taken from the current period. The
formula for the Paasche is given below in equation 3:

=
t, 0

P .Q
P .Q
i

i,t

i,0

i, t

i, t

100

(3)

So, whereas the Laspeyres calculates what expenditure in the

current period would be if the quantity consumed were the same as
in the base period (a pure price effect), the Paasche calculates the
expenditure needed to buy current year quantities, and is expressed
as a percentage of what the expenditure would have been in the
base period if the quantity consumed had been at current levels.
Examples of a Paasche price index, both unchained and chained,
are shown below in Table 4 and Table 5 using the same data as in
previous examples.
Unlike the Laspeyres, in the case of the Paasche, chain-linking
has the effect of reducing the index. This is because growth is
not calculated as a percentage of expenditure in the base period
but instead is backward-looking. Therefore, the effect of the high
increases in the price of X, and substitution from Y to X, is less
pronounced when the index is chained together. Put another way,
there is an incremental weighting effect, that is, the shift in the
weights (quantities) is more pronounced between the starting and

Table 4
Example of a weighted Paasche index

Current

Current
expenditure
expenditure: using base year
prices: P0*Qt
Time (t) P(X)
P(Y)
Q(X)
Q(Y)
Pt*Qt
2000
2001
2002
2003
2004
2005
2006

10
14
19
28
40
44
56

5
7
6
7
8
9
11

30
32
38
43
47
53
60

60
60
868
620
55
1,052
655
52
1,568
690
50
2,280
720
50
2,782
780
45
3,855
825

Index:
base
(2000)
100
140
161
227
317
357
467

Table 5
Example of a weighted chain-linked Paasche index

Current

Current
expenditure
expenditure: using base year
Time (t) P(X)
P(Y)
Q(X)
Q(Y)
Pt*Qt prices: Pt-1*Qt
2000
2001
2002
2003
2004
2005
2006

10
14
19
28
40
44
56

5
7
6
7
8
9
11

30
32
38
43
47
53
60

Index:
chainlinked

60
60
868
620
55
1,052
917
52
1,568
1,129
50
2,280
1,666
50
2,782
2,520
45
3,855
3,045

100
140
161
223
305
337
427

Office for National Statistics

55

Methods explained

Economic & Labour Market Review | Vol 1 | No 3 | March 2007

current period than it is between the previous and current period,

causing the index to increase more slowly.

Table 7
Example of a weighted chain-linked Fisher index

The choice of whether to use a Laspeyres or Paasche is fairly

arbitrary. The decision will probably make very little difference
to the final index unless there has been a substantial change to
the weighting of the variable, as there has been in the examples
above. Therefore, the decision tends to be based on practicalities.
To construct a Paasche index, as explained above, weights for the
actual year of the series, or current weights, are required. Such data
are often unavailable. Therefore, the Laspeyres tends to be used, as
weights from a previous period are more readily available. Also,
using a Paasche index means that the denominator changes every
year, so different years can only be compared with the base year and
not with each other.

Time (t)
Laspeyres
Paasche
Product

An example of a Paasche index produced by ONS is the implied

GDP deflator, which is also chain-linked.

Fisher
The Fisher index, or the Fisher ideal index, is a form of compromise
between the Laspeyres and Paasche. Its formula is a geometric mean
of the Laspeyres and Paasche, as shown in equations 4 and 5. In
general, the Laspeyres is always greater than or equal to the Fisher,
and the Paasche is always less than or equal to the Fisher (6):

F
t, 0

P .Q P .Q

P .Q P .Q

i,t

i,0

i, t

i, t

)(

i,t

i,0

i, 0

i, t

100

(4)

(5)

Laspeyres Fisher Paasche

(6)

It should be noted that the reason the inequality given in equation

6 does not hold in the above examples is that the numbers used for
price and quantity suggest a perverse demand relationship. That is,
in general, if there is an increase in the price ratio (PX/PY) of the two
goods, we would expect the quantity ratio (QX/QY) to decrease. This
is not the case in the hypothetical numerical examples given.
In general, the Fisher is used when prices and quantities in the base
and observation periods are substantially different. Examples of a
Fisher index, both unchained and chained, are provided below in
Table 6 and Table 7, again using the same data as used previously.

Office for National Statistics

10,000
19,600
25,076
47,969
89,429
109,032
174,555

100
140
158
219
299
330
418

The Fisher index is termed ideal because it gets around the practical
problem of time reversal. This is particularly relevant when the
data are seasonal and was encountered when attempts were made
to construct a quarterly Laspeyres version of QALI. Basically, what
happens is that when growth in the variable increases, but in a
subsequent period decreases, the index fails to decrease all the way
back and remains at a higher level. Superlative2 indices, such as
the Fisher, and also the Trnqvist, which is discussed in the next
section, overcome this problem and this is the reason why they are
sometimes preferred.

100
140
161
227
317
357
467

Trnqvist
The main feature of the Trnqvist index is that the weight used is
an average of the weight in the current and base period. Therefore,
like the Paasche, it tends to be used on historic data sets, as current
period information is needed to weight the series. Another difference
is that it is calculated geometrically rather than arithmetically. The
formula for a Trnqvist index is shown below in equations 7 and 8,
and examples of both unchained and chained Trnqvist indices are
shown below in Table 8 and Table 9 (the price and quantity data are
the same as those used previously, but are omitted to save space).

( )

Pi,t
I = P
i,0
i

T
t ,0

(w

+wi,0
)
2

i,t

(7)

where:
Pi,tQi,t

wi,t =

Time (t)
Laspeyres
Paasche
Product

56

100
140
161
223
305
337
427

(8)

P ,Q

Table 6
Example of a weighted Fisher index

100
140
155
210
280
310
390

100
140
156
215
293
324
409

Another convenient property is that multiplying a Fisher price and

Fisher quantity index produces a Fisher expenditure index.

IFt, 0 = ILt, 0 IPt, 0

2000
2001
2002
2003
2004
2005
2006

2000
2001
2002
2003
2004
2005
2006

Fisher:
square root
of product

10,000
19,600
24,895
47,722
88,667
110,567
182,236

Fisher:
square root
of product
100
140
158
218
298
333
427

it

i,t

For ease of composition, the Trnqvist is often written as:

It,0T = exp

(
i

)1n (P )]

wi,t+wi,0
2

Pi,t

i,0

(9)

Economic & Labour Market Review | Vol 1 | No 3 | March 2007

Methods explained

Table 8
Example of a weighted Trnqvist index



Relative
Share

change
of Good
Time (t)
in P(X)
X
2000
2001
2002
2003
2004
2005
2006

1.0
1.4
1.9
2.8
4.0
4.4
5.6

0.50
0.52
0.69
0.77
0.82
0.84
0.87

Share of
X
averaged
Relative
over both
change
Share of
in P(Y)
Good Y
periods
X^share
0.50
0.51
0.59
0.63
0.66
0.67
0.69

1.00
1.19
1.46
1.92
2.50
2.69
3.26

1.00
1.40
1.20
1.40
1.60
1.80
2.20

0.50
0.48
0.31
0.23
0.18
0.16
0.13

Share of
Y
averaged
over both
(X^share)
periods
Y^share
*(Y^share)
0.50
0.49
0.41
0.37
0.34
0.33
0.31

1.00
1.18
1.08
1.13
1.17
1.21
1.28

Index:
base
(2000)

1.00
1.40
1.58
2.17
2.94
3.27
4.18

100
140
158
217
294
327
418

Share of
Y
averaged
over both
(X^share)
periods
Y^share
*(Y^share)

Index:
chainlinked

Table 9
Example of a weighted chain-linked Trnqvist index



Relative
Share

change
of Good
Time (t)
in P(X)
X
2000
2001
2002
2003
2004
2005
2006

1.00
1.40
1.36
1.47
1.43
1.10
1.27

0.50
0.52
0.69
0.77
0.82
0.84
0.87

Share of
X
averaged
Relative
over both
change
Share of
in P(Y)
Good Y
periods
X^share
0.50
0.51
0.60
0.73
0.80
0.83
0.85

1.00
1.19
1.20
1.33
1.33
1.08
1.23

The Trnqvist has a number of useful features. As well as removing

the problem of time reversal, it can also be used to show the
contribution of each component to aggregate growth because it
is log-linear. Therefore, the contributions can be broken down
into additive form. Another feature is they correspond to translog
production and cost functions and are hence widely used in
empirical work. For instance, using a standard Cobb-Douglas
production function where:
Y = KL(1-)

(10)

and:
Y = output,
K = capital,
L = labour
and (1-) = income shares of K and L, respectively
In general, the income shares are assumed to be constant. However,
the use of Trnqvist indices means these are more flexible and can
change according to the returns to labour or capital.

Conclusion

1.00
1.40
0.86
1.17
1.14
1.13
1.22

0.50
0.48
0.31
0.23
0.18
0.16
0.13

0.25
0.49
0.40
0.27
0.20
0.17
0.15

1.00
1.18
0.94
1.04
1.03
1.02
1.03

1.00
1.40
1.13
1.38
1.37
1.10
1.27

100
140
158
219
299
330
417

Notes
1 Technically the RPI is not a true Laspeyres index, but it is a very
close approximation. For further detail and reasons consult the
Retail Price Index Technical Manual, available at www.statistics.
gov.uk/statbase/Product.asp?vlnk=2328
2 A superlative index is one that uses more information in its
construction than a base index, and is more flexible. The OECD
definition is as follows: Superlative indices are price or quantity
indices that are exact for a flexible aggregator. A flexible
aggregator is a second-order approximation to an arbitrary
production, cost, utility or distance function. Exactness implies
that a particular index number can be directly derived from a
specific flexible aggregator.
References
Caswell F (1995) Success in Statistics Third Edition, Coventry University
Harper M (1991) Statistics Sixth Edition (1991), Longman Group UK
Limited
Robjohns J (2006) Methodology Notes: Annual chain-linking
Economic Trends 630, pp 258 and at www.statistics.gov.uk/cci/article.
asp?ID=1554

There are, therefore, various forms of indices and their construction

affects the result and how they should be interpreted. The choice of
which to use tends to depend on what is being compared over time,
the timeliness of the available data, how much flexibility is needed
and what its final use is, for example, is it going to be used as an
input into another piece of analysis which requires an index of a
certain form.

Office for National Statistics

57